Assigning trimester-specific exposure windows in a birth cohort study

Trimester-specific exposure assignment sounds like a bookkeeping problem until you sit down with a cohort's pregnancy records and realize gestational age isn't one number, it's an estimate built on a dating method, and the window boundaries you pick will shift which pollution days land in which trimester for every participant.
Pick the dating method before you pick the cutoffs
Most cohorts have two candidate anchors: last menstrual period (LMP) and first-trimester ultrasound. ACOG guidance favors ultrasound dating when the two disagree by more than the standard thresholds (roughly 5-7 days in the first trimester, widening later), because LMP recall is noisy and cycles aren't always 28 days. If your cohort collected both, document which one you used as the primary anchor and run the trimester assignment once with each, then look at how much the window dates move. In our experience pulling this kind of comparison together, the shift matters most for participants near a trimester boundary, which is exactly where exposure misclassification does the most damage.
Standard trimester definitions run first trimester at 0-13 completed weeks, second at 14-26, and third at 27 weeks to delivery. Some groups use 0-12/13-27/28-delivery instead. Neither is wrong, but mixing definitions across a multi-site cohort or across your own prior publications creates headaches during meta-analysis, so write the cutoff choice into your protocol before the first participant's dates get processed.
Handle preterm birth and delivery date separately
A birth at 34 weeks doesn't get a truncated third trimester, it just has a shorter one. Don't backfill exposure days past the actual delivery date to make the window match a term pregnancy's length. This sounds obvious written out, but it's a common bug when trimester windows are built with a fixed day count instead of being anchored to estimated conception date and the actual delivery date for that participant. Build the window boundaries as functions of individual dates, not as a template applied uniformly.
The harder case is the participant delivered before her second or third trimester even started under the standard cutoffs. For an extremely preterm birth, you may only have a first and partial second trimester. Decide ahead of time whether you drop that participant from trimester-specific models, keep her in the pregnancy-average model only, or flag incomplete windows as a covariate. Any of these is defensible. Not having a rule is what gets flagged in review.
Match the window to the right exposure surface
Once you have start and end dates for each trimester, the next step is pulling daily pollutant concentrations for the residential address on file and averaging across each window. This is where the exposure data itself becomes the limiting factor. A monthly or even weekly PM2.5 product forces you to apportion partial periods across trimester boundaries with interpolation assumptions that add noise you didn't need. A daily surface lets you sum or average the exact days in a window without guessing what happened on the days you don't have.
Address history matters here too. When a participant moved during pregnancy and you have geocoded addresses with move dates, assign each trimester's exposure to the home address she occupied during that specific window. Cohorts that only capture a single address lose this distinction entirely, which is worth flagging as a limitation rather than pretending the exposure is more precise than it is.
The nearest regulatory monitor often sits miles from where a participant lived, which is fine for a city-level average but weak for a trimester-specific estimate tied to one home address. If your study area has monitor gaps or you need exposure at the resolution of an individual address rather than a county, a daily satellite-derived surface built for a defined study area gets you concentration estimates at the address level for every day of the pregnancy window, which is the unit trimester assignment actually needs.
Write the trimester definition, the dating method, and the address-matching rule into your protocol as three separate, explicit decisions. Reviewers will ask about all three, and having the answer already in your methods section saves a revise-and-resubmit round. If daily, address-level PM2.5 coverage over your cohort's exact study area is the piece you're missing, that's worth a closer look.